Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,6 +1,5 @@
|
|
| 1 |
import os
|
| 2 |
import sys
|
| 3 |
-
import subprocess
|
| 4 |
from threading import Thread
|
| 5 |
from typing import Iterable
|
| 6 |
from huggingface_hub import snapshot_download
|
|
@@ -21,14 +20,6 @@ from gradio.themes.utils import colors, fonts, sizes
|
|
| 21 |
|
| 22 |
# --- Theme and CSS Definition ---
|
| 23 |
|
| 24 |
-
# Attempt to install flash-attn
|
| 25 |
-
try:
|
| 26 |
-
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, check=True, shell=True)
|
| 27 |
-
except subprocess.CalledProcessError as e:
|
| 28 |
-
print(f"Error installing flash-attn: {e}")
|
| 29 |
-
print("Continuing without flash-attn.")
|
| 30 |
-
|
| 31 |
-
|
| 32 |
colors.steel_blue = colors.Color(
|
| 33 |
name="steel_blue",
|
| 34 |
c50="#EBF3F8",
|
|
@@ -158,14 +149,12 @@ model_m = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
|
| 158 |
torch_dtype=torch.float16
|
| 159 |
).to(device).eval()
|
| 160 |
|
| 161 |
-
import flash_attn_2_cuda as flash_attn_gpu
|
| 162 |
-
|
| 163 |
# Load Dots.OCR from the local, patched directory
|
| 164 |
MODEL_PATH_D = model_path_d_local
|
| 165 |
processor_d = AutoProcessor.from_pretrained(MODEL_PATH_D, trust_remote_code=True)
|
| 166 |
model_d = AutoModelForCausalLM.from_pretrained(
|
| 167 |
MODEL_PATH_D,
|
| 168 |
-
attn_implementation="
|
| 169 |
torch_dtype=torch.bfloat16,
|
| 170 |
device_map="auto",
|
| 171 |
trust_remote_code=True
|
|
|
|
| 1 |
import os
|
| 2 |
import sys
|
|
|
|
| 3 |
from threading import Thread
|
| 4 |
from typing import Iterable
|
| 5 |
from huggingface_hub import snapshot_download
|
|
|
|
| 20 |
|
| 21 |
# --- Theme and CSS Definition ---
|
| 22 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
colors.steel_blue = colors.Color(
|
| 24 |
name="steel_blue",
|
| 25 |
c50="#EBF3F8",
|
|
|
|
| 149 |
torch_dtype=torch.float16
|
| 150 |
).to(device).eval()
|
| 151 |
|
|
|
|
|
|
|
| 152 |
# Load Dots.OCR from the local, patched directory
|
| 153 |
MODEL_PATH_D = model_path_d_local
|
| 154 |
processor_d = AutoProcessor.from_pretrained(MODEL_PATH_D, trust_remote_code=True)
|
| 155 |
model_d = AutoModelForCausalLM.from_pretrained(
|
| 156 |
MODEL_PATH_D,
|
| 157 |
+
attn_implementation="eager",
|
| 158 |
torch_dtype=torch.bfloat16,
|
| 159 |
device_map="auto",
|
| 160 |
trust_remote_code=True
|